Must Read Papers on Deep Learning for the Image Signal Processor (ISP)

Overview

The Image Signal Processor (ISP) is a fundamental processing pipeline in modern cameras and smartphones. The ISP is responsible for mapping RAW sensor data to a visually pleasing RGB image for end-user consumption. Typical stages in the pipleline include, but are not limited to, denoising, demosaicing, high-dynamic range compression, colour mapping, sharpening. The actual makeup of ISP pipelines are typically very closely guarded commercial secrets.

Recently there has been great interest in the machine learning and computer vision research communities towards investigating how the aforementioned operations in the ISP could be replaced by deep neural networks. This website presents an ever-growing hand-curated selection of papers that leverage deep learning to greatly improve the image quality resulting from individual components of the ISP pipleline, and in a brave recent development, those papers that attempt to model the entire pipeline with a single deep neural network.

Browse Papers by Tag

Deep Learning Denoising End-to-End Kernel Prediction Network TIP

Contributing

To add a new paper to the website simply create a markdown file and open a pull request in GitHub by following these instructions for contributing.